fix debug flag and kernel launch parameter querying

seeder-research · May 9, 2022 · 48470e1 · 48470e1
1 parent 66d53c3
commit 48470e1
Show file tree

Hide file tree

Showing 5 changed files with 51 additions and 26 deletions.
diff --git a/cmd/uMagNUS/main.go b/cmd/uMagNUS/main.go
@@ -26,6 +26,9 @@ func main() {
 	log.SetPrefix("")
 	log.SetFlags(0)
 
+	opencl.Synchronous = *engine.Flag_sync
+	opencl.Debug = *engine.Flag_debug
+
 	// Check flag and initialize engine
 	if len(*engine.Flag_gpulist) > 0 {
 		var gpu_arr []int
@@ -63,7 +66,6 @@ func main() {
 		}
 	}
 
-	opencl.Synchronous = *engine.Flag_sync
 	if *engine.Flag_version {
 		printVersion()
 	}

diff --git a/cmd/uMagNUS64/main.go b/cmd/uMagNUS64/main.go
@@ -26,6 +26,9 @@ func main() {
 	log.SetPrefix("")
 	log.SetFlags(0)
 
+	opencl.Synchronous = *engine.Flag_sync
+	opencl.Debug = *engine.Flag_debug
+
 	// Check flag and initialize engine
 	if len(*engine.Flag_gpulist) > 0 {
 		var gpu_arr []int
@@ -63,7 +66,6 @@ func main() {
 		}
 	}
 
-	opencl.Synchronous = *engine.Flag_sync
 	if *engine.Flag_version {
 		printVersion()
 	}

diff --git a/opencl/init.go b/opencl/init.go
@@ -25,7 +25,7 @@ var (
 	GPUInfo      string                    // Human-readable GPU description
 	GPUList      []GPU                     // List of GPUs available
 	Synchronous  bool                      // for debug: synchronize command queue at every kernel launch
-	Debug        bool                      // for debug: synchronize command queue after every kernel launch
+	Debug        = false                   // for debug: synchronize command queue after every kernel launch
 	ClPlatforms  []*cl.Platform            // list of platforms available
 	ClPlatform   *cl.Platform              // platform the global OpenCL context is attached to
 	ClDevices    []*cl.Device              // list of devices global OpenCL context may be associated with
@@ -209,13 +209,13 @@ func Init(gpu int) {
 	inRegExp := regexp.MustCompile("(?i)intel")
 	adRegExp0 := regexp.MustCompile("(?i)amd")
 	adRegExp1 := regexp.MustCompile("(?i)micro device")
-	if chk0 := nvRegExp.Match([]byte(GPUInfo)); chk0 {
+	if chk0 := nvRegExp.Match([]byte(PlatformInfo)); chk0 {
 		GPUVend = 1
 	} else {
-		if chk1 := inRegExp.Match([]byte(GPUInfo)); chk1 {
+		if chk1 := inRegExp.Match([]byte(PlatformInfo)); chk1 {
 			GPUVend = 2
 		} else {
-			chk2, chk3 := adRegExp0.Match([]byte(GPUInfo)), adRegExp1.Match([]byte(GPUInfo))
+			chk2, chk3 := adRegExp0.Match([]byte(PlatformInfo)), adRegExp1.Match([]byte(PlatformInfo))
 			if (chk2 == true) || (chk3 == true) {
 				GPUVend = 3
 			} else {
@@ -224,13 +224,13 @@ func Init(gpu int) {
 		}
 	}
 	ClMaxWGNum = ClCUnits
+	ClTotalPE = ClWGSize[2] * ClCUnits
 	if GPUVend == 1 { // Nvidia
-		ClTotalPE = ClWGSize[2] * ClCUnits
 		if ClMaxWGSize > ClTotalPE {
-			ClMaxWGNum = ClTotalPE / ClMaxWGSize
-		} else {
 			ClMaxWGNum = 1
 			ClMaxWGSize = ClTotalPE
+		} else {
+			ClMaxWGNum = ClTotalPE / ClMaxWGSize
 		}
 	}
 	if GPUVend == 2 { // Intel
@@ -252,6 +252,16 @@ func Init(gpu int) {
 	reduceintcfg.Grid[0] = ClTotalPE
 	reduceintcfg.Block[0] = ClPrefWGSz
 
+	if Debug {
+		fmt.Printf("    PlatformInfo: \n%+v \n", PlatformInfo)
+		fmt.Printf("    GPUInfo: \n%+v \n", GPUInfo)
+		fmt.Printf("    GPUVend: %+v \n", GPUVend)
+		fmt.Printf("    ClCUnits: %+v ; ClWGSize = %+v \n", ClCUnits, ClWGSize)
+		fmt.Printf("    ClTotalPE = %+v \n", ClTotalPE)
+		fmt.Printf("    ClMaxWGSize = %+v ; ClMaxWGNum = %+v \n", ClMaxWGSize, ClMaxWGNum)
+		fmt.Printf("    ClPrefWGSz = %+v \n", ClPrefWGSz)
+	}
+
 	data.EnableGPU(memFree, memFree, MemCpy, MemCpyDtoH, MemCpyHtoD)
 
 }

diff --git a/opencl/opencl.go b/opencl/opencl.go
@@ -1,10 +1,11 @@
 package opencl
 
 import (
-	"log"
+	"fmt"
 	"unsafe"
 
 	cl "github.com/seeder-research/uMagNUS/cl"
+	util "github.com/seeder-research/uMagNUS/util"
 )
 
 // Type size in bytes
@@ -18,15 +19,15 @@ const (
 // Assumes kernel arguments set prior to launch
 func LaunchKernel(kernname string, gridDim, workDim []int, events []*cl.Event) *cl.Event {
 	if KernList[kernname] == nil {
-		log.Panic("Kernel " + kernname + " does not exist!")
+		util.Fatal("Kernel " + kernname + " does not exist!")
 		return nil
 	}
 	if Debug {
-		log.Printf("Launching kernel: %+v with Grid = %+v and Block = %+v \n", kernname, gridDim, workDim)
+		fmt.Printf("Launching kernel: %+v with Grid = %+v and Block = %+v \n", kernname, gridDim, workDim)
 	}
 	KernEvent, err := ClCmdQueue.EnqueueNDRangeKernel(KernList[kernname], nil, gridDim, workDim, events)
 	if err != nil {
-		log.Fatal(err)
+		util.Fatal(err)
 		return nil
 	} else {
 		return KernEvent
@@ -35,31 +36,31 @@ func LaunchKernel(kernname string, gridDim, workDim []int, events []*cl.Event) *
 
 func SetKernelArgWrapper(kernname string, index int, arg interface{}) {
 	if KernList[kernname] == nil {
-		log.Panic("Kernel " + kernname + " does not exist!")
+		util.Fatal("Kernel " + kernname + " does not exist!")
 	}
 	switch val := arg.(type) {
 	default:
 		if err := KernList[kernname].SetArg(index, val); err != nil {
-			log.Fatal(err)
+			util.Fatal(err)
 		}
 	case unsafe.Pointer:
 		memBufHandle, flag := arg.(unsafe.Pointer)
 		if memBufHandle == unsafe.Pointer(uintptr(0)) {
 			if err := KernList[kernname].SetArgUnsafe(index, 8, memBufHandle); err != nil {
-				log.Fatal(err)
+				util.Fatal(err)
 			}
 		} else {
 			if flag {
 				if err := KernList[kernname].SetArg(index, (*cl.MemObject)(memBufHandle)); err != nil {
-					log.Fatal(err)
+					util.Fatal(err)
 				}
 			} else {
-				log.Fatal("Unable to change argument type to *cl.MemObject")
+				util.Fatal("Unable to change argument type to *cl.MemObject")
 			}
 		}
 	case int:
 		if err := KernList[kernname].SetArg(index, (int32)(val)); err != nil {
-			log.Fatal(err)
+			util.Fatal(err)
 		}
 	}
 }
diff --git a/opencl64/init.go b/opencl64/init.go
@@ -25,7 +25,7 @@ var (
 	GPUInfo      string                    // Human-readable GPU description
 	GPUList      []GPU                     // List of GPUs available
 	Synchronous  bool                      // for debug: synchronize command queue at every kernel launch
-	Debug        bool                      // for debug: synchronize command queue after every kernel launch
+	Debug        = false                   // for debug: synchronize command queue after every kernel launch
 	ClPlatforms  []*cl.Platform            // list of platforms available
 	ClPlatform   *cl.Platform              // platform the global OpenCL context is attached to
 	ClDevices    []*cl.Device              // list of devices global OpenCL context may be associated with
@@ -212,13 +212,13 @@ func Init(gpu int) {
 	inRegExp := regexp.MustCompile("(?i)intel")
 	adRegExp0 := regexp.MustCompile("(?i)amd")
 	adRegExp1 := regexp.MustCompile("(?i)micro device")
-	if chk0 := nvRegExp.Match([]byte(GPUInfo)); chk0 {
+	if chk0 := nvRegExp.Match([]byte(PlatformInfo)); chk0 {
 		GPUVend = 1
 	} else {
-		if chk1 := inRegExp.Match([]byte(GPUInfo)); chk1 {
+		if chk1 := inRegExp.Match([]byte(PlatformInfo)); chk1 {
 			GPUVend = 2
 		} else {
-			chk2, chk3 := adRegExp0.Match([]byte(GPUInfo)), adRegExp1.Match([]byte(GPUInfo))
+			chk2, chk3 := adRegExp0.Match([]byte(PlatformInfo)), adRegExp1.Match([]byte(PlatformInfo))
 			if (chk2 == true) || (chk3 == true) {
 				GPUVend = 3
 			} else {
@@ -227,13 +227,13 @@ func Init(gpu int) {
 		}
 	}
 	ClMaxWGNum = ClCUnits
+	ClTotalPE = ClWGSize[2] * ClCUnits
 	if GPUVend == 1 { // Nvidia
-		ClTotalPE = ClWGSize[2] * ClCUnits
 		if ClMaxWGSize > ClTotalPE {
-			ClMaxWGNum = ClTotalPE / ClMaxWGSize
-		} else {
 			ClMaxWGNum = 1
 			ClMaxWGSize = ClTotalPE
+		} else {
+			ClMaxWGNum = ClTotalPE / ClMaxWGSize
 		}
 	}
 	if GPUVend == 2 { // Intel
@@ -255,6 +255,16 @@ func Init(gpu int) {
 	reduceintcfg.Grid[0] = ClTotalPE
 	reduceintcfg.Block[0] = ClPrefWGSz
 
+	if Debug {
+		fmt.Printf("    PlatformInfo: \n%+v \n", PlatformInfo)
+		fmt.Printf("    GPUInfo: \n%+v \n", GPUInfo)
+		fmt.Printf("    GPUVend: %+v \n", GPUVend)
+		fmt.Printf("    ClCUnits: %+v ; ClWGSize = %+v \n", ClCUnits, ClWGSize)
+		fmt.Printf("    ClTotalPE = %+v \n", ClTotalPE)
+		fmt.Printf("    ClMaxWGSize = %+v ; ClMaxWGNum = %+v \n", ClMaxWGSize, ClMaxWGNum)
+		fmt.Printf("    ClPrefWGSz = %+v \n", ClPrefWGSz)
+	}
+
 	data.EnableGPU(memFree, memFree, MemCpy, MemCpyDtoH, MemCpyHtoD)
 
 }